Unverified50% confidenceFactExact time
VoxCPM2 adopts a tokenizer-free architecture, removing both the text tokenizer and the acoustic tokenizer stages, letting the model complete the mapping from text to speech in a continuous representation space.
1
Sources
50%
Confidence
Long-term
Relevance
10/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedContinuous diffusion language models map discrete tokens to a continuous embedding space, perform the noising and denoising process in that space, and finally decode back to discrete tokens.70% similarUnverifiedThe lexical-prosodic unitization and chunking method performs pooling only on consecutive word units to preserve event identity and magnitude in both the text and speech modalities.68% similarUnverifiedSelf-supervised speech encoders compress linguistic and paralinguistic information into the same shared, entangled representation space.67% similarVerifiedNeural audio codecs such as EnCodec and DAC compress continuous audio into discrete token sequences, enabling speech to be directly processed by autoregressive language models just like text tokens.67% similarVerifiedWhisper's architecture adopts an encoder-decoder structure, where the encoder transforms the mel spectrogram of audio into high-dimensional feature representations, and the decoder generates text output word by word in an autoregressive manner.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1089326API
curl https://kongchang.com/api/v1/knowledge/claims/1089326MCP
get_claim(id=1089326)